Utilizing an Ensemble Framework for Real-Time Spatiotemporal Data Streams Concept Drift Handling in Crime Classification

Abstract

The number of systems and devices broadcasting spatiotemporal data has recently significantly increased. Streaming data analytics provides the foundation of various spatiotemporal data services and functions. The non-stationary characteristics of these platforms and the constantly altering trends of the spatiotemporal data streams present concept drift issues for spatiotemporal data analytics. As a result, when concept drift occurs, it harms the model. The model's performance will eventually decline. The learning algorithms need the proper adaptive techniques to deal with concept drift on the spatiotemporal data streams with accurate predictions. This paper proposes an average weighted performance ensemble model (AWPEM). The AWPEM framework is for drift adaptation for spatiotemporal crime prediction. Compared to state-of-the-art approaches, the results experiment on two crime datasets demonstrated the proposed AWPEM method's efficiency in concept drift detections and accurate spatiotemporal crime prediction was higher.